Shamar / ANN-decompiler
To prove that any "artificial neural network" is just a statistically programmed (virtual) machine whose model software is a derivative work of the source dataset used during its "training", we provide a small suite of tools to assemble and program such machines and a decompiler that reconstruct the source dataset from the cryptic matrices that constitute the software executed by them. Finally we test the suite on the classic MNIST dataset and compare the decompiled dataset with the original one.
Artificial Intelligence, Machine Learning, Artificial Neural Networks, Microsoft, GitHub Copilot, Python, Statistical Programming, Vector Mapping Machine
Despite the hazy anthropomorphic language and the unprofessional narrative, in the last few years the field of Informatics that goes under the names of "Artificial Intelligence" and "Machine Learning" established a long record of self-appointed successes. Behind the hype and the sci-fi depictions of marvelous futures where virtual humans will explore the Universe on behalf of a mostly extincted humanity spread to slow down any effective solution that could save the lifes of billions of real people, there are few misunderstood (and often misused) techniques that actually bring some innovation to the table.
The most iconic of such innovations is improperly named "artificial neural network", after a wrong model of the human brain that inspired their creation back in the sixties. On top of this class of tools, composed and programmed in a multitude of ways, several useful services are being commoditized by large corporations with access to huge dataset, money and computational power.
